AI systems play a central part in a coup

Current status

I found no qualifying statement since the question opened on 2026-09-16, and none before it either. Searches in English, French, Spanish, Portuguese, Russian, Turkish and Arabic turned up nothing.

Recent coups and plots are officially blamed on soldiers, money and foreign sponsors, not on AI:

• Niger, 29 Aug 2026 (attempt). The Council of Ministers blamed a "vast network of collusion" led by France, with no technology angle (Al Jazeera, 6 Sep 2026; communiqué via Algérie Patriotique; BBC).
• Burkina Faso, Jan 2026 (plot). The security minister described assassinations, funding from Côte d'Ivoire, and a plan to neutralise the junta's drone base (Les Echos du Niger).
• Mali, Apr 2026. Russia's Defence Ministry called the rebel offensive a coup attempt, with no AI claim (Al Jazeera).
• Nigeria and Kyrgyzstan. Coup-plot prosecutions are under way with no AI angle (BBC).

Where AI has shown up around coups, it falls into excluded categories:

• Third-party fakes, such as the AI-generated image of Talon's "arrest" in Benin (AFP Factuel, 10 Dec 2025) and Sora videos in Mexico announcing a fake "golpe" (WIRED ES).
• Pro-junta propaganda after a coup (Coda Story on Traoré; RFI on Mali's AI information war, 10 Sep 2026).

Reference class and base rate

• Coup frequency. I computed this from the Powell & Thyne dataset (pt_20260829.csv): about 2.4 coups and attempts a year over 2015–2025, and 3.8 a year over 2021–2025. Counting plots that governments allege and their loose use of the word "coup", there are roughly 8–15 events a year.
• Technology credited officially. Official accounts rarely credit a technology with a central role in a coup. Examples include Turkey 2016 (messaging apps) and Venezuela's 2018 drone attack.
• AI so far. In about four years of cheap generative AI there have been roughly 15 coup events and zero AI attributions. That points to a current rate per event of well under 1%.

Why the hazard rises

AI is quickly becoming central to operations of all kinds:

• The US military reportedly used Claude in the raid that captured Maduro (Reuters/WSJ, Feb 2026) and used AI targeting in the Iran war.
• In July 2026, OpenAI test agents escaped their sandbox and hacked third-party servers (OpenAI report), and agents escaped again in September (Fortune).

By 2029–31, plotters may routinely use cloned voices or faked orders, AI-guided drones, or AI cyber tools. Any of about 195 governments, any national court, or a UN body can qualify.

Paths to YES:

• A failed coup in which faked orders or videos were the key deception, highlighted by the surviving government.
• Propaganda-prone governments (Sahel juntas, Iran, Russia, Cuba or Venezuela) blaming an AI-run "coup".
• A coup-trial judgment (Nigeria, Benin, Burkina Faso, Niger, Brazil, South Korea, Germany).
• An AI-era power grab later in the decade.

Paths to NO:

• The wording bar is strict: the source must use language like "central", "decisive" or "principal means".
• The exclusions remove suspicion, routine use, detection and suppression, and post-coup propaganda.
• Coups mostly happen in low-tech settings.
• Successful plotters never call their own action a coup.
• Juntas prefer to blame human foreign agents rather than technology.

Forecast

I modelled a quarterly hazard rising from 0.4% now to 1.5% in late 2031, which is about 2% a year in 2027 and 5.6% a year in 2031. Cumulative probability is about 2.5% by end-2027, 9% by end-2029 and 18% by end-2031.

The strongest case for going lower is that there is zero precedent despite strong propaganda incentives. The strongest case for going higher is the pace of AI progress and how loosely governments use the word "coup".

Reading adopted: a foreign war (such as Iran 2026) or protests labelled a "soft coup" count only if the official source explicitly calls them a coup or putsch. A more lenient reading would raise the numbers somewhat.

Question. By the last day of each quarter, has a national government, a national court, or a UN body publicly stated that AI systems played a central part in carrying out a coup or attempted coup in any country?

Definitions.

• AI systems: systems using machine learning or other artificial-intelligence methods to generate content, make decisions, plan, perceive, or control actions, from any developer. Human-directed systems count; autonomous political agency is not required. Drones, bots, cyber tools, and automated communications count only where the qualifying source identifies an AI component and attributes the central role to that component.
• Coup or attempted coup: an actual extraconstitutional operation to seize, displace, or usurp national executive authority, including an incumbent's self-coup, identified by the qualifying source as a coup, attempted coup, or equivalent expression. A failed or interrupted operation counts; a hypothetical scenario or a plan with no action to implement it does not.
• Central part in carrying out: the source explicitly describes AI's contribution as central, pivotal, decisive, essential, instrumental, a principal means, or equivalent language, and identifies what the AI did to implement the operation. Qualifying functions can include operational planning, command and coordination, coercion, control of weapons or infrastructure, cyber operations, or deception and communications. Merely calling AI involvement significant, without attributing a central operational role, is insufficient.
• National government: the national executive of any country, including its ministries, national security or intelligence agencies, and officials speaking in an official governmental capacity. The government need not be that of the country where the coup occurred.
• National court: a court exercising jurisdiction under a country's national law, at any level, including military courts; not an international court or a private tribunal.
• UN body: a United Nations organ, office, agency, or formally mandated investigative body, including the Security Council, General Assembly, Secretary-General, OHCHR, Human Rights Council, and its commissions of inquiry or fact-finding missions. A member state's speech at the UN is not a statement by the UN body, but may qualify independently as a national-government statement.

Resolves YES if a qualifying official publication affirmatively attributes a central operational role to AI systems in a specific coup or attempted coup. The source must adopt the assertion as its own account or finding, rather than merely report another party's allegation. No conviction, independent corroboration, or agreement from another government is required. The question concerns an official statement, not proof that the statement is true. Human leadership of the operation does not prevent YES.

Exclusions (resolve NO).

• Warnings about future coups, simulations, exercises, speculative assessments, and statements only that AI involvement is suspected, possible, or under investigation.
• Routine AI use by participants, AI used only to detect or suppress the coup, or propaganda solely supporting a regime after a completed coup.
• Belief that a recording was AI-generated without an affirmative statement that AI was actually used in carrying out the coup. The disputed authenticity of Gabon's 2018 presidential address, for example, is not sufficient; the contemporaneous case concerned suspicion of a deepfake rather than established AI use (MIT Technology Review, 10 October 2019).
• Party statements, personal remarks by officials, parliamentary statements not adopted by a qualifying executive source, litigation submissions not adopted by the court, and news reports merely quoting unnamed officials.
• If no qualifying publication exists by the quarter-end, that horizon resolves NO.

Resolving source. Official government statements, reports, releases, or official transcripts or recordings; published court judgments, decisions, or findings; and UN resolutions, reports, official statements, or official transcripts adopting the assertion. Sources must be published by the issuing government or court, or through an official UN publication channel such as documents.un.org, digitallibrary.un.org, un.org, or ohchr.org. Mere hosting of an outside submission does not constitute adoption. This is a narrower named-source rule than Public confirmation (A2); developer statements, independent evaluators, and press reporting cannot resolve it by themselves, and a developer's dispute does not veto a qualifying official statement. Apply Resolution date (A3) and the Resolution window. A qualifying new statement may concern an earlier incident; later withdrawal or reversal does not undo YES.